The Impact of Eyebrows Region on Deep Face Recognition

Abstract

We investigate the importance of the information contained in the area of the eyebrows for deep face recognition. An isotropic 2D Gaussian low-pass filter of varying bandwidth is used to remove discriminative information in the probe and reference images gradually. We measure the recognition performance of two deep learning-based face recognition systems as a function of the bandwidth of the low-pass filter applied to the eyebrows. Methods are tested on the frontal face and high-resolution images from the PUT database. The results showed that even though the eyebrows are important for recognition, deep learning still works well on facial images with removed eyebrows. Furthermore, we found that the discriminative information provided by eyebrows comes from their shapes, not their textures.